A method for ground texture navigation

Abstract

The present invention belongs to the field of navigation in indoor venues, specifically a method for ground texture navigation. It includes the following steps: a camera collects images in real time and stores and sends them to an image preprocessing module for processing; determines whether the current navigation vehicle's driving speed is less than a speed threshold, and if so, executes a low-speed posture positioning algorithm; otherwise, executes a high-speed posture positioning algorithm; obtains the absolute two-dimensional posture of the current camera relative to the initial position through a low-speed posture positioning algorithm or executes a high-speed posture positioning algorithm; and positions the navigation vehicle. In the present invention, two applicable methods correspond to different usage scenarios, and can be used alone to cope with a single usage scenario, or can be used together, and switched in real time to cope with different usage requirements in different scenarios. Among them, the low-speed posture positioning algorithm is suitable for high-precision, general speed usage scenarios, and the high-speed posture positioning algorithm is suitable for non-high-precision, faster moving speed usage scenarios.

Description

Technical Field

[0001] The invention belongs to the field of indoor site navigation, and in particular to a ground texture navigation method. Background Art

[0002] With the rapid development of artificial intelligence (AI) technology and the widespread use of mobile robots in various fields, positioning systems, as a key component of robotic systems, are gaining increasing attention. From outdoor GPS to high-precision indoor laser, vision, and ultrasonic navigation, precise robot navigation in diverse environments is gaining increasing attention. Common mobile robot navigation methods include magnetic stripe navigation, inertial navigation, laser navigation, and vision navigation.

[0003] Both magnetic stripe navigation and QR code navigation in visual navigation have high positioning accuracy, but when the magnetic stripe and QR code are aged and damaged, positioning loss is likely to occur, and neither safety nor robustness can be guaranteed; when using laser navigation, according to its positioning principle, mobile robots cannot be used in a wide environment; in addition, traditional visual SLAM navigation has more stringent environmental requirements and cannot accurately locate in environments with weak textures, repetitive textures, and drastically changing lighting conditions. Summary of the Invention

[0004] The present invention aims to provide a method for ground texture navigation, a two-dimensional visual odometry navigation solution for weak-texture and repetitive-texture surfaces on construction sites. This solution installs a fisheye (or wide-angle) monocular camera and a high-intensity light below the chassis of a mobile platform, maintaining a constant height above the ground. The ground texture captured by the camera is analyzed and registered in real time, thereby effectively estimating the real-time two-dimensional position of the camera. This solution can solve the problem of mobile robot positioning on flat surfaces with repetitive or weak textures, reduce safety risks, and improve production efficiency.

[0005] The technical solution adopted by the present invention to achieve the above-mentioned purpose is: a ground texture navigation method, comprising the following steps:

[0006] 1) The camera collects and stores images in real time; and sends the collected original image of the current frame and the previous frame to the image preprocessing module for processing;

[0007] 2) Determine whether the current navigation vehicle speed is less than the set speed value. If so, execute the low-speed posture positioning algorithm; otherwise, execute the high-speed posture positioning algorithm;

[0008] 3) When executing the low-speed pose positioning algorithm, the image preprocessing module preprocesses the current frame original image and the previous frame original image, and sends the preprocessed current frame image and the previous frame image to the phase correlation preprocessing module for processing, and obtains the logarithmic polar coordinate transformed image of the two frames and sends it to the first phase correlation module; at the same time, the preprocessed current frame image and the previous frame image are sent to the second phase correlation module for processing;

[0009] The first phase correlation module processes the two frames of log-polar coordinate transformed images to obtain the relative rotation angle and scaling ratio of the two frames of images, and sends them to the second phase correlation module;

[0010] The second phase correlation module transforms the images of the two frames according to the relative rotation angle and scaling ratio of the two frames, as well as the logarithmic polar coordinates of the two frames, to obtain the rotated and scaled image of the current frame and the previous frame, thereby obtaining the absolute two-dimensional pose of the current camera relative to the initial position;

[0011] 4) When executing the high-speed pose positioning algorithm, the image preprocessing module performs processing to obtain the preprocessed and cropped current frame image and the previous frame image, and sends them to the third phase correlation module for phase correlation processing to obtain the translation amount of the current frame image and the translation amount of the previous frame image after the two frames are cropped, as well as the response map of the images at the same position of the two frames;

[0012] Each translation amount and the corresponding response value in the response graph are sent to the data fusion module for processing to obtain the absolute two-dimensional pose of the current camera relative to the initial position;

[0013] 5) Positioning the navigation vehicle based on the absolute two-dimensional pose of the current camera relative to the initial position obtained in step 3) or step 4).

[0014] In step 3), the image preprocessing module processes the current frame original image and the previous frame original image, specifically:

[0015] Distortion correction and smoothing filtering are performed on the current frame original image and the previous frame original image respectively to eliminate the distortion of the two frames and reduce noise;

[0016] The preprocessed current frame image and the previous frame image are cropped to obtain a cropped current frame image and a cropped previous frame image respectively.

[0017] In step 3), the pre-processed current frame image and the previous frame image are sent to the phase correlation pre-processing module for processing to obtain the logarithmic polar coordinate transformation images of the two frames of images, specifically:

[0018] Fourier transform is performed on the cropped current frame image and the cropped previous frame image respectively to obtain corresponding amplitude spectra; amplitude maps corresponding to the current frame image and the previous frame image are obtained according to the amplitude spectra, and the amplitude maps of the two frames of images are subjected to high-pass filtering for noise reduction to obtain two noise-reduced amplitude maps;

[0019] Perform log-polar coordinate transformation on the two amplitude images after noise reduction respectively to obtain log-polar coordinate transformed images of the two frames of images, and send them to the first phase correlation module.

[0020] In step 3), the first phase correlation module processes the two frames of log-polar coordinate transformed images to obtain the relative rotation angle and scaling ratio of the two frames of images, specifically:

[0021] The first phase correlation module performs Fourier transform and phase correlation processing on the two frames of log-polar coordinate transformed images, and then inverse transforms them to obtain a response map. By finding the position of the maximum peak in the response map, the relative rotation angle and scaling ratio of the two frames of images are determined.

[0022] In step 3), the second phase correlation module obtains the absolute two-dimensional pose of the current camera relative to the initial position based on the relative rotation angle and scaling ratio of the two frames of image and the logarithmic polar coordinate transformation images of the two frames of image, specifically:

[0023] The second phase correlation module processes the current frame image according to the scale and rotation to obtain an image with the same scale and posture as the previous frame, and uses this image and the previous frame image to perform a second phase correlation process, also obtaining a response graph. The position (X, Y) of the maximum peak in the response graph is found, which is the position change between adjacent frames;

[0024] Through coordinate system transformation, the absolute two-dimensional pose (X, Y, θ) of the current camera relative to the initial frame position can be obtained.

[0025] In step 4), the pre-processed and cropped current frame image and the previous frame image are obtained respectively, specifically:

[0026] The image preprocessing module performs dedistortion processing on the current frame original image and the previous frame original image, and performs image cropping processing on the two processed frames respectively; the two frames are cropped and divided into two equal-sized square images, L image and R image, and L' image and R' image respectively;

[0027] The L image and L' image, as well as the R image and R' image at the same position in the two frames of images are respectively sent to the third phase correlation module for processing.

[0028] In the step 4), the phase correlation module is sent to the third phase correlation module for phase correlation processing to obtain the translation amount of each cropped current frame image and the translation amount of the previous frame image as well as the response value of the current frame image and the response value of the previous frame image, specifically:

[0029] Phase correlation processing is performed on the L image and L' image, as well as the R image and R' image at the same position in the two frames of images, to obtain the response map of the images at the same position in the two frames of images;

[0030] By finding the maximum peak position of the response graph, the translation amounts δx1 and δy1 of the L' image relative to the L image, the translation amounts δx2 and δy2 of the R' image relative to the R image, and the corresponding response values ​​res1 and res2 are obtained and sent to the data fusion module.

[0031] In step 4), each translation amount and each response value are sent to the data fusion module for processing to obtain the absolute two-dimensional pose of the current camera relative to the initial position, specifically:

[0032] According to each translation amount, the inverse tangent function is used to obtain the rotation angle of the current frame image relative to the previous frame image, that is:

[0033] According to each translation amount, the inverse tangent function is used to obtain the rotation angle of the current frame image relative to the previous frame image, that is:

[0034]

[0035] Among them, D V is the horizontal distance between the rotated images L' and R', D H is the vertical distance between the rotated image L' and image R', D LR is the distance between the center points of image L and image R;

[0036] Use res1 as the weight to calculate the displacement of L image and L' image, and use res2 as the weight to calculate the weighted average of the displacement of R image and R' image, that is:

[0037]

[0038]

[0039] As the displacement X, Y of the current frame relative to the previous frame; then the posture change of the current frame image relative to the previous frame image is obtained;

[0040] According to the coordinate system transformation, the absolute two-dimensional pose (X, Y, θ) of the current camera relative to the initial position can be obtained.

[0041] A mobile vehicle based on a ground texture navigation method comprises: a camera, a light source and a sunshade arranged under a mobile chassis of the navigation vehicle;

[0042] The illumination area of ​​the light source is the same as the field of view of the camera, and a light shield is provided around the camera and the light source to prevent external interference light from entering the field of view of the camera;

[0043] The camera is a wide-angle camera or a fisheye camera, and the optical axis of the camera is perpendicular to the ground.

[0044] A navigation system based on a ground texture navigation method includes: an image preprocessing module, a phase correlation preprocessing module, a first phase correlation module, a second phase correlation module, a third phase correlation module, and a data fusion module;

[0045] The image preprocessing module has an input end connected to the camera and an output end connected to the phase correlation preprocessing module, the second phase correlation module, and the third phase correlation module, respectively, and is used to receive the current frame original image and the previous frame original image sent by the camera, and perform preprocessing, and send the result image obtained by preprocessing to the third phase correlation module, or the phase correlation preprocessing module and the second phase correlation module;

[0046] The phase correlation preprocessing module is used to process the result image obtained by the image preprocessing module to obtain a logarithmic polar coordinate transformed image of the two frames of images, and send it to the first phase correlation module; the first phase correlation module is connected to the phase correlation preprocessing module and the second phase correlation module respectively, and is used to receive the logarithmic polar coordinate transformed image of the two frames of images sent by the phase correlation preprocessing module, process it, obtain the relative rotation angle and scaling ratio of the two frames of images, and send it to the second phase correlation module;

[0047] The second phase correlation module is connected to the first phase correlation module and is used to receive the result image preprocessed by the image preprocessing module, and process the result image in combination with the relative rotation angle and scaling ratio of the two frames of images sent by the first phase correlation module, that is, to obtain the rotated and scaled image of the current frame and the previous frame of image, thereby obtaining the absolute two-dimensional pose of the current camera relative to the initial position;

[0048] The third phase correlation module is connected to the data fusion module, and is used to receive the result image obtained by preprocessing for processing, obtain the translation amount of the current frame image and the translation amount of the previous frame image, and the response map of the images at the same position of the two frames, and send them to the data fusion module for processing;

[0049] The data fusion module is used to process the translation amount of the current frame image and the translation amount of the previous frame image and the response map of the two frames of images at the same position to obtain the absolute two-dimensional pose of the current camera relative to the initial position.

[0050] The present invention has the following beneficial effects and advantages:

[0051] 1. The visual texture navigation estimation solution of the present invention, consisting of a fisheye camera (or wide-angle camera), a strong light, and a cloth, is easy to install and disassemble, and can also be easily moved to other platforms or sites. The larger the camera's wide-angle angle, the closer it can be installed to the ground, and the requirements for the platform's mechanical structure are relatively relaxed.

[0052] 2. The present invention can be effectively used on various flat surfaces with weak textures and repeated textures, including but not limited to floors, marble, cement floors, etc., effectively solving the problem of inability to perform visual positioning in construction sites with repeated textures that are common nowadays.

[0053] 3. In this invention, two applicable methods are used for different usage scenarios. They can be used alone to address a single usage scenario, or they can be used together and switched in real time to meet the different usage requirements in different scenarios. Among them, the low-speed pose positioning algorithm is suitable for high-precision, medium-speed usage scenarios, while the high-speed pose positioning algorithm is suitable for non-high-precision, high-speed usage scenarios.

[0054] 4. Compared with the common navigation methods used in other factories, the method adopted by the present invention can work without changing the site environment, while achieving a certain degree of accuracy, so that the mobile platform can rely solely on vision for positioning and navigation. In addition, compared with other cameras that require a significant change in the mechanical structure of existing equipment when installed, the device designed in this solution is more convenient to install, requiring a flat site and maintaining a certain distance from the ground. Compared with other ground texture navigation solutions based on Fourier Merlin, the present invention proposes Solution 2, which significantly improves the calculation speed while slightly reducing the accuracy, and alleviates the computing pressure of the computing equipment to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a structural diagram of the navigation vehicle of the present invention;

[0056] Figure 2 The low-speed posture positioning algorithm and system principle diagram of the present invention;

[0057] Figure 3 The high-speed posture positioning algorithm and system principle diagram of the present invention;

[0058] Figure 4 Obtain the rotation angle principle diagram for the present invention. DETAILED DESCRIPTION

[0059] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0060] This solution consists of two parts: hardware and software. Figure 1 As shown, it is a schematic diagram of the structure of the navigation vehicle of the present invention. The hardware part is the navigation vehicle. The navigation vehicle uses a camera, which is composed of a camera, a light source and a cloth. The camera is a wide-angle camera or a fisheye camera. The light source in this embodiment uses a strong light ring. When the mobile robot light body is installed, the cloth uses a blackout cloth or a blackout plate. The cloth is used to surround the surrounding environment of the camera so that external light does not enter the camera's field of view. At the same time, the strong light ring is turned on around the camera to make uniform light spread throughout the camera's field of view, so that the brightness of each frame of the image obtained by the camera remains constant.

[0061] like Figure 2 He Ru Figure 3 As shown, it is a flow chart of the method principle of the system of the present invention; the system of low-speed posture positioning algorithm, such as Figure 2 As shown, it is mainly divided into the following four modules:

[0062] (1) Image preprocessing module:

[0063] The main purpose is to pre-process the original image captured by the camera to eliminate distortion and reduce noise. At the same time, the image is cropped (because the Fourier transform needs to process square images) for subsequent use.

[0064] (2) Phase correlation method preprocessing module:

[0065] The cropped image is subjected to the first Fourier transform to obtain the corresponding amplitude map, which is then subjected to high-pass filtering for noise reduction, and finally to a logarithmic-polar coordinate transformation (this process is also called the Fourier-Mellin transform).

[0066] (3) First phase correlation module:

[0067] The logarithmic polar coordinate images obtained from two adjacent frames are subjected to Fourier transform and phase correlation, and then inversely transformed to obtain a response map. By finding the position of the maximum peak in the response map, the relative rotation angle and scaling ratio (theta, scale) of the two frames are determined.

[0068] (4) Second phase correlation module:

[0069] The current frame image is processed according to scale and rotation to obtain an image with the same scale and posture as the previous frame, and the new image and the previous frame image are used for the second phase correlation registration. The response map is also obtained, and the position (x, y) of the maximum peak in the response map is found, which is the position change of the adjacent frames.

[0070] Finally, through coordinate system transformation, the absolute two-dimensional pose (X, Y, θ) of the current camera relative to the initial frame position can be obtained.

[0071] The system structure of the high-speed posture positioning algorithm is as follows Figure 3As shown in the figure, when the mobile robot's movement speed is greater than the set speed, the high-speed posture positioning algorithm strategy will be used to estimate the real-time position of the mobile robot. It is mainly divided into the following three modules:

[0072] (1) Image preprocessing module:

[0073] The image is dedistorted and cropped into two equal-sized square images, L and R.

[0074] (2) The third phase correlation module:

[0075] Phase correlation processing is performed on the images at the same position in the current frame and the previous frame respectively, and the translation amounts Δx1, Δy1, Δx2, Δy2 corresponding to LR and the corresponding response values ​​res1 and res2 are obtained by finding the maximum peak position of the response graph.

[0076] (3) Data fusion module:

[0077] According to the difference in displacement of the two parts of the image, the rotation angle thata can be calculated;

[0078] like Figures 2 and 3 As shown, a ground texture navigation method of the present invention includes the following steps:

[0079] 1) The camera collects and stores images in real time; and sends the collected original image of the current frame and the previous frame to the image preprocessing module for processing;

[0080] 2) Determine whether the current navigation vehicle speed is less than the set speed value. If so, execute the low-speed posture positioning algorithm; otherwise, execute the high-speed posture positioning algorithm;

[0081] 3) When executing the low-speed pose positioning algorithm, the image preprocessing module preprocesses the current frame original image and the previous frame original image, and sends the preprocessed current frame image and the previous frame image to the phase correlation preprocessing module for processing, and obtains the logarithmic polar coordinate transformed image of the two frames and sends it to the first phase correlation module; at the same time, the preprocessed current frame image and the previous frame image are sent to the second phase correlation module for processing;

[0082] The first phase correlation module processes the two frames of log-polar coordinate transformed images to obtain the relative rotation angle and scaling ratio of the two frames of images, and sends them to the second phase correlation module;

[0083] The second phase correlation module transforms the images of the two frames according to the relative rotation angle and scaling ratio of the two frames, as well as the logarithmic polar coordinates of the two frames, to obtain the rotated and scaled image of the current frame and the previous frame, thereby obtaining the absolute two-dimensional pose of the current camera relative to the initial position;

[0084] 4) When executing the high-speed pose positioning algorithm, the image preprocessing module performs processing to obtain the preprocessed and cropped current frame image and the previous frame image, and sends them to the third phase correlation module for phase correlation processing to obtain the translation amount of the current frame image and the translation amount of the previous frame image after the two frames are cropped, as well as the response map of the images at the same position of the two frames;

[0085] Each translation amount and the corresponding response value in the response graph are sent to the data fusion module for processing to obtain the absolute two-dimensional pose of the current camera relative to the initial position;

[0086] 5) Positioning the navigation vehicle based on the absolute two-dimensional pose of the current camera relative to the initial position obtained in step 3) or step 4).

[0087] Among them, Figure 2 As shown in the figure, the specific steps of using the low-speed posture positioning algorithm are:

[0088] The image preprocessing module performs distortion correction and smoothing filtering on the current frame original image and the previous frame original image respectively to eliminate the distortion of the two frames and reduce noise;

[0089] The pre-processed current frame image and the previous frame image are cropped to obtain the cropped current frame image and the cropped previous frame image respectively. The images are then sent to the phase correlation pre-processing module to obtain the log-polar coordinate transformation images of the two frames.

[0090] Fourier transform is performed on the cropped current frame image and the cropped previous frame image respectively to obtain corresponding amplitude spectra; amplitude maps corresponding to the current frame image and the previous frame image are obtained according to the amplitude spectra, and the amplitude maps of the two frames of images are subjected to high-pass filtering for noise reduction to obtain two noise-reduced amplitude maps;

[0091] Perform log-polar coordinate transformation on the two amplitude images after noise reduction respectively to obtain log-polar coordinate transformed images of the two frames of images, and send them to the first phase correlation module.

[0092] The first phase correlation module performs Fourier transform and phase correlation processing on the two frames of log-polar coordinate transformed images, and then inverse transforms them to obtain a response map. By finding the position of the maximum peak in the response map, the relative rotation angle and scaling ratio of the two frames of images are determined.

[0093] The second phase correlation module processes the current frame image according to the scale and rotation to obtain an image with the same scale and posture as the previous frame, and uses this image and the previous frame image to perform a second phase correlation process, also obtaining a response graph. The position (X, Y) of the maximum peak in the response graph is found, which is the position change between adjacent frames;

[0094] Through coordinate system transformation, the absolute two-dimensional pose (X, Y, θ) of the current camera relative to the initial frame position can be obtained.

[0095] like Figure 3 As shown in FIG, the flow chart of the high-speed posture positioning algorithm system of the present invention is as follows:

[0096] The image preprocessing module performs dedistortion processing on the current frame original image and the previous frame original image, and performs image cropping processing on the two processed frames respectively; the two frames are cropped and divided into two equal-sized square images, L image and R image, and L' image and R' image respectively;

[0097] The L image and L' image, as well as the R image and R' image at the same position in the two frames of images are respectively sent to the third phase correlation module for processing.

[0098] The third phase correlation module performs phase correlation processing on the L image and L' image, as well as the R image and R' image at the same position in the two frames of images, to obtain a response map of the images at the same position in the two frames of images;

[0099] By finding the maximum peak position of the response graph, the translation amounts δx1 and δy1 of the L' image relative to the L image, the translation amounts δx2 and δy2 of the R' image relative to the R image, and the corresponding response values ​​res1 and res2 are obtained and sent to the data fusion module.

[0100] like Figure 4 As shown in FIG, the principle diagram of obtaining the rotation angle of the present invention is shown. The data fusion module obtains the rotation angle of the current frame image relative to the previous frame image through the inverse tangent function according to each translation amount, that is:

[0101] According to each translation amount, the inverse tangent function is used to obtain the rotation angle of the current frame image relative to the previous frame image, that is:

[0102]

[0103] Among them, D V is the horizontal distance between the rotated images L' and R', D H is the vertical distance between the rotated image L' and image R', D LR is the distance between the center points of image L and image R;

[0104] Use res1 as the weight to calculate the displacement of L image and L' image, and use res2 as the weight to calculate the weighted average of the displacement of R image and R' image, that is:

[0105]

[0106]

[0107] As the displacement X, Y of the current frame relative to the previous frame; then the posture change of the current frame image relative to the previous frame image is obtained;

[0108] According to the coordinate system transformation, the absolute two-dimensional pose (X, Y, θ) of the current camera relative to the initial position can be obtained.

[0109] Compared to the low-speed pose positioning algorithm, the high-speed pose positioning algorithm does not calculate the angle and scaling values ​​using the first phase correlation method. Instead, it assumes that the calculation speed is fast enough that the small rotation angle between the two frames is not enough to affect the phase correlation method's calculation of displacement. First, the image is cropped, and the current frame and the previous frame are divided into two square images (L, R) (L', R') with smaller resolution. Phase correlation is performed on the L, R, L', and R' of the previous and next frames to calculate the displacement δx1, δy1, δx2, and δy2. The small rotation angle is then calculated using the inverse tangent function. Finally, the displacement X and Y are calculated using a weighted average.

[0110] For visual navigation on surfaces with weak or repetitive textures, the present invention uses a transform-domain approach to analyze and register ground textures, fusing the registered information from consecutive adjacent frames into a visual odometry. This visual odometry is sufficiently accurate for mobile robots to locate and navigate in various locations. It can also be combined with other sensor data fusion (such as inertial navigation) for even more precise positioning, enabling tasks requiring even higher precision.

[0111] The above description is only an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modification, equivalent replacement, improvement, expansion, etc. made within the spirit and principle of the present invention are included in the scope of protection of the present invention.

Claims

1. A method for ground texture navigation, characterized in that: The following steps are involved: 1) The camera collects and stores images in real time; and sends the collected original image of the current frame and the previous frame to the image preprocessing module for processing; 2) Determine whether the current navigation vehicle speed is less than the set speed value. If so, execute the low-speed posture positioning algorithm; otherwise, execute the high-speed posture positioning algorithm; 3) When executing the low-speed pose positioning algorithm, the image preprocessing module preprocesses the current frame original image and the previous frame original image, and then sends the preprocessed current frame image and the previous frame image to the phase correlation preprocessing module for processing. The obtained logarithmic polar coordinate transformed image of the two frames is sent to the first phase correlation module; at the same time, the preprocessed current frame image and the previous frame image are sent to the second phase correlation module for processing; The first phase correlation module processes the two frames of log-polar coordinate transformed images to obtain the relative rotation angle and scaling ratio of the two frames of images, and sends them to the second phase correlation module; The second phase correlation module transforms the images of the two frames according to the relative rotation angle and scaling ratio of the two frames, as well as the logarithmic polar coordinates of the two frames, to obtain the rotated and scaled image of the current frame and the previous frame, thereby obtaining the absolute two-dimensional pose of the current camera relative to the initial position; 4) When executing the high-speed pose positioning algorithm, the image preprocessing module performs processing to obtain the preprocessed and cropped current frame image and the previous frame image, and sends them to the third phase correlation module for phase correlation processing to obtain the translation amount of the current frame image and the translation amount of the previous frame image after the two frames are cropped, as well as the response map of the images at the same position of the two frames; Each translation amount and the corresponding response value in the response graph are sent to the data fusion module for processing to obtain the absolute two-dimensional pose of the current camera relative to the initial position; 5) Positioning the navigation vehicle based on the absolute two-dimensional pose of the current camera relative to the initial position obtained in step 3) or step 4).

2. A method for ground texture navigation according to claim 1, characterized in that: In step 3), the image pre-processing module processes the current frame original image and the previous frame original image, specifically: Distortion correction and smoothing filtering are performed on the current frame original image and the previous frame original image respectively to eliminate the distortion of the two frames and reduce noise; The preprocessed current frame image and the previous frame image are cropped to obtain a cropped current frame image and a cropped previous frame image respectively.

3. The method for ground texture navigation according to claim 1, characterized in that: In step 3), the pre-processed current frame image and the previous frame image are sent to the phase correlation pre-processing module for processing to obtain the logarithmic polar coordinate transformation images of the two frames of images, specifically: Fourier transform is performed on the cropped current frame image and the cropped previous frame image respectively to obtain corresponding amplitude spectra; amplitude maps corresponding to the current frame image and the previous frame image are obtained according to the amplitude spectra, and the amplitude maps of the two frames of images are subjected to high-pass filtering for noise reduction to obtain two noise-reduced amplitude maps; Perform log-polar coordinate transformation on the two amplitude images after noise reduction respectively to obtain log-polar coordinate transformed images of the two frames of images, and send them to the first phase correlation module.

4. The method for ground texture navigation according to claim 1, characterized in that: In step 3), the first phase correlation module processes the two frames of log-polar coordinate transformed images to obtain the relative rotation angle and scaling ratio of the two frames of images, specifically: The first phase correlation module performs Fourier transform and phase correlation processing on the two frames of log-polar coordinate transformed images, and then inverse transforms to obtain a response map. By finding the position of the maximum peak in the response map, the relative rotation angle of the two frames of images is determined. and scaling.

5. The method for ground texture navigation according to claim 1, characterized in that: In step 3), the second phase correlation module obtains the absolute two-dimensional pose of the current camera relative to the initial position based on the relative rotation angle and scaling ratio of the two frames of image and the logarithmic polar coordinate transformation images of the two frames of image, specifically: The second phase correlation module processes the current frame image according to the scale and rotation to obtain an image with the same scale and posture as the previous frame, and uses this image and the previous frame image to perform phase correlation processing, and also obtains a response graph, and finds the position of the maximum peak in the response graph ( ) is the position change of adjacent frames; Through coordinate system conversion, the absolute two-dimensional position (X, Y, ).

6. The method for ground texture navigation according to claim 1, characterized in that: In step 4), the pre-processed and cropped current frame image and previous frame image are obtained respectively, specifically: The image preprocessing module performs dedistortion processing on the current frame original image and the previous frame original image, and performs image cropping processing on the two processed frames respectively; the two frames are cropped and divided into two equal-sized square images, L image and R image, and L' image and R' image respectively; The L image and L' image, as well as the R image and R' image at the same position in the two frames of images are respectively sent to the third phase correlation module for processing.

7. The method for ground texture navigation according to claim 1, characterized in that: In step 4), the phase correlation processing is performed on the third phase correlation module to obtain the translation amount of each cropped current frame image and the translation amount of the previous frame image, as well as the response value of the current frame image and the response value of the previous frame image, specifically: Phase correlation processing is performed on the L image and L' image, as well as the R image and R' image at the same position in the two frames of images, to obtain the response map of the images at the same position in the two frames of images; By finding the maximum peak position of the response graph, the translation of the L' image relative to the L image is obtained. , the translation of the R' image relative to the R image , and the corresponding response values ​​res1 and res2, and sent to the data fusion module.

8. A method for ground texture navigation according to claim 1 or 7, characterized in that: In step 4), each translation amount and each response value are sent to the data fusion module for processing to obtain the absolute two-dimensional pose of the current camera relative to the initial position, specifically: According to each translation amount, the inverse tangent function is used to obtain the rotation angle of the current frame image relative to the previous frame image, that is: ； in, is the horizontal distance between the rotated images L' and R', is the vertical distance between the rotated image L' and image R', is the distance between the center points of image L and image R; Use res1 as the weight to calculate the displacement of L image and L' image, and use res2 as the weight to calculate the weighted average of the displacement of R image and R' image, that is: ； ； As the displacement of the current frame relative to the previous frame ; Then obtain the pose change of the current frame image relative to the previous frame image; According to the coordinate system conversion, the absolute two-dimensional pose (X, Y, ).

9. A mobile vehicle according to any one of claims 1 to 8, characterized in that: include: A camera, a light source, and a sunshade are arranged under the mobile chassis of the navigation vehicle; The illumination area of ​​the light source is the same as the field of view of the camera, and a light shield is provided around the camera and the light source to prevent external interference light from entering the field of view of the camera; The camera is a wide-angle camera or a fisheye camera, and the optical axis of the camera is perpendicular to the ground.

10. A navigation system based on a ground texture navigation method according to any one of claims 1 to 8, characterized in that: include: An image preprocessing module, a phase correlation preprocessing module, a first phase correlation module, a second phase correlation module, a third phase correlation module, and a data fusion module; The image preprocessing module has an input end connected to the camera and an output end connected to the phase correlation preprocessing module, the second phase correlation module, and the third phase correlation module, respectively, and is used to receive the current frame original image and the previous frame original image sent by the camera, and perform preprocessing, and send the result image obtained by preprocessing to the third phase correlation module, or the phase correlation preprocessing module and the second phase correlation module; The phase correlation preprocessing module is used to process the result image obtained by the image preprocessing module to obtain a logarithmic polar coordinate transformation image of the two frames of images, and send it to the first phase correlation module; The first phase correlation module is connected to the phase correlation preprocessing module and the second phase correlation module respectively, and is used to receive the log-polar coordinate transformed images of the two frames of images sent by the phase correlation preprocessing module, process them, obtain the relative rotation angle and scaling ratio of the two frames of images, and send them to the second phase correlation module; The second phase correlation module is connected to the first phase correlation module and is used to receive the result image preprocessed by the image preprocessing module, and process the result image in combination with the relative rotation angle and scaling ratio of the two frames of images sent by the first phase correlation module, that is, to obtain the rotated and scaled image of the current frame and the previous frame of image, thereby obtaining the absolute two-dimensional pose of the current camera relative to the initial position; The third phase correlation module is connected to the data fusion module, and is used to receive the result image obtained by preprocessing for processing, obtain the translation amount of the current frame image and the translation amount of the previous frame image, and the response map of the images at the same position of the two frames, and send them to the data fusion module for processing; The data fusion module is used to process the translation amount of the current frame image and the translation amount of the previous frame image and the response map of the two frames of images at the same position to obtain the absolute two-dimensional pose of the current camera relative to the initial position.

Images